Abstract: A computer-implemented method for learning a policy for selection of an associative topic, which can be used in a dialog system, is described. The method includes obtaining a policy base that indicates a topic transition from a source topic to a destination topic and a short-term reward for the topic transition, by analyzing data from a corpus. The short-term reward may be defined as probability of associating a positive response. The method also includes calculating an expected long-term reward for the topic transition using the short-term reward for the topic transition with taking into account a discounted reward for a subsequent topic transition. The method further includes generating a policy using the policy base and the expected long-term reward for the topic transition. The policy indicates selection of the destination topic for the source topic as an associative topic for a current topic.

Abstract: Computations in Artificial neural networks (ANNs) are accomplished using simple processing units, called neurons, with data embodied by the connections between neurons, called synapses, and by the strength of these connections, the synaptic weights. Crossbar arrays may be used to represent one layer of the ANN with Non-Volatile Memory (NVM) elements at each crosspoint, where the conductance of the NVM elements may be used to encode the synaptic weights, and a highly parallel current summation on the array achieves a weighted sum operation that is representative of the values of the output neurons. A method is outlined to transfer such neuron values from the outputs of one array to the inputs of a second array with no need for global clock synchronization, irrespective of the distances between the arrays, and to use such values at the next array, and/or to convert such values into digital bits at the next array.

Abstract: Systems are provided for implementing a hardware accelerator. The hardware accelerator emulate a stochastic neural network, and includes a first memristor crossbar array, and a second memristor crossbar array. The first memristor crossbar array can be programmed to calculate node values of the neural network. The nodes values can be calculated in accordance with rules to reduce an energy function associated with the neural network. The second memristor crossbar array is coupled to the first memristor crossbar array and programmed to introduce noise signals into the neural network. The noise signals can be introduced such that the energy function associated with the neural network converges towards a global minimum and modifies the calculated node values.

Abstract: A computer-implemented method for learning a policy for selection of an associative topic, which can be used in a dialog system, is described. The method includes obtaining a policy base that indicates a topic transition from a source topic to a destination topic and a short-term reward for the topic transition, by analyzing data from a corpus. The short-term reward may be defined as probability of associating a positive response. The method also includes calculating an expected long-term reward for the topic transition using the short-term reward for the topic transition with taking into account a discounted reward for a subsequent topic transition. The method further includes generating a policy using the policy base and the expected long-term reward for the topic transition. The policy indicates selection of the destination topic for the source topic as an associative topic for a current topic.

Abstract: Methods and systems are disclosed for data retrieval, from databases to clients, in an environment requiring runtime authorization. In response to a request for T data records, a learning module provides a prediction R of a suitable number of data records to retrieve from a database. Following retrieval of R records or record identifiers, authorization is sought from an authorization service, resulting in A of the records being authorized. The A authorized records are returned to the requesting client, and, if more records are needed, T is decremented and the cycle is repeated. A performance notification is provided to the learning module for training, with respect to providing values of prediction R. The performance notification can be based on a measure of authorization service performance, the number A of authorized records, latency, communication or resource costs, a measure of resource congestion, or other parameters. Variants are disclosed.

Abstract: Devices and methods for learning and/or predicting the self-reported pain improvement levels of osteoarthritis (OA) patients are provided. A device or apparatus can include a processor and a machine-readable medium in operable communication with the processor and having stored thereon an algorithm and a unique set of features. The algorithm and set of features can enable building one or more models that learn the self-reported pain improvement levels of OA patients.

Abstract: Systems and methods for training a multitask network is disclosed. In one aspect, training the multitask network includes determining a gradient norm of a single-task loss adjusted by a task weight for each task, with respect to network weights of the multitask network, and a relative training rate for the task based on the single-task loss for the task. Subsequently, a gradient loss function, comprising a difference between (1) the determined gradient norm for each task and (2) a corresponding target gradient norm, can be determined. An updated task weight for the task can be determined and used in the next iteration of training the multitask network, using a gradient of the gradient loss function with respect to the task weight for the task.

Abstract: Methods, apparatuses, and embodiments related to a technique for monitoring construction of a structure. In an example, a robot with a sensor, such as a LIDAR device, enters a building and obtains sensor readings of the building. The sensor data is analyzed and components related to the building are identified. The components are mapped to corresponding components of an architect's three dimensional design of the building, and the installation of the components is checked for accuracy. When a discrepancy above a certain threshold is detected, an error is flagged and project managers are notified. Construction progress updates do not give credit for completed construction that includes an error, resulting in improved accuracy progress updates and corresponding improved accuracy for project schedule and cost estimates.

Abstract: This disclosure relates to learning with transformed data such as determining multiple training samples from multiple data samples. Each of the multiple data samples comprises one or more feature values and a label that classifies that data sample. A processor determines each of the multiple training samples by randomly selecting a subset of the multiple data samples, and combining the feature values of the data samples of the subset based on the label of each of the data samples of the subset. Since the training samples are combinations of randomly chosen data samples, the training samples can be provided to third parties without disclosing the actual training data. This is an advantage over existing methods in cases where the data is confidential and should therefore not be shared with a learner of a classifier, for example.

Abstract: A method of making a mirror for use with extreme ultraviolet (EUV) or X-ray radiation is disclosed. The method includes: a) providing an optical element having a curved mirror surface, wherein the curved mirror surface comprises localized defects that degrade performance of the curved mirror surface; b) spin-coating the curved mirror surface with a material to cover at least some of the defects; and c) curing the spin-coated material on the curved mirror surface to reduce the number of defects and improve the performance of the curved mirror surface. Also disclosed is a mirror made by the method.

Abstract: Methods and apparatus to perform multi-level hierarchical demographic classification are disclosed.

Abstract: Systems and methods of implementing a more efficient and less resource-intensive CNN are disclosed herein. In particular, applications of CNN in the analog domain using Sampled Analog Technology (SAT) methods are disclosed. Using a CNN design with SAT results in lower power usage and faster operation as compared to a CNN design with digital logic and memory. The lower power usage of a CNN design with SAT can allow for sensor devices that also detect features at very low power for isolated operation.

Abstract: A computer-implemented method applies labels to unlabeled public data for use by a global model. One or more processors train one or more local machine learning models with local private data to create one or more trained models. Processor(s) generate a label for each of the local private data using the one or more trained models, where each label describes the local private data, and then apply the label to unlabeled public data to create labeled public data. One or more processors then input the labeled public data into a global model that uses the public data.

Abstract: Described are a system, method, and computer program product for optimizing a predictive condition classification model and automatically enacting reactive measures based thereon. The method includes receiving event data representative of a plurality of events. The method also includes receiving the predictive condition classification model configured to categorize each event as satisfying a condition or not. The predictive condition classification model is configured to order the plurality of events by likelihood of satisfying the condition. The method includes generating a performance evaluation dataset and plotting data configured to cause a visual display to represent at least two model performance metrics of the performance evaluation dataset on a same output plot. The method includes automatically rejecting a top percent of the plurality of events for suspected satisfaction of the condition, determined at least partially from a customized rejection algorithm or a preset rejection algorithm.

Abstract: Systems and methods for ordering software applications in a computing environment. The methods involve: presenting user-selectable icons for launching a plurality of software applications in a graphical user interface in accordance with a first order; performing a machine-learning algorithm to determine a weighting value for each software application of the plurality of software applications based on information specifying at least one aspect of a software launch request and at least one aspect of a first user's current circumstance; determining a second order in which the user-selectable icons should be presented in the graphical user interface based on the weighting values determined for the software applications; and dynamically modifying the graphical user interface to present the user-selectable icons in the second order which is different from the first order.

Abstract: Methods, computer program products, and systems are presented. The methods include, for instance: obtaining communication data streams, extracting data relevant to a point of view of a user, and generating a point of view record in a knowledge base that may be utilized by another user communicating with the user.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for adversarial training of a neural network. One of the methods includes obtaining a plurality of training inputs; and training the neural network on each of the training inputs, comprising, for each of the training inputs: processing the training input using the neural network to determine a neural network output for the training input; applying a perturbation to the training input to generate an adversarial perturbation of the training input; processing the adversarial perturbation of the training input using the neural network to determine a neural network output for the adversarial perturbation; and adjusting the current values of the parameters of the neural network by performing an iteration of a neural network training procedure to optimize an adversarial objective function.

Abstract: Embodiments of the present disclosure provide a substrate measuring device in a lithography projection apparatus that provides multiple light sources having different wavelengths. In some embodiments, a lithography projection apparatus includes a substrate measuring system disposed proximate to a substrate stage, the substrate measuring system further including an emitter including multiple light sources configured to provide multiple beams of light, each of at least some of the multiple beams of light having a different wavelength, at least one optical fiber, wherein each of respective portions of the at least one optical fiber is configured to pass a respective one of the multiple beams of light, and a receiver positioned to collected light emitted from the emitter and reflected off of a substrate disposed on the substrate stage.

Abstract: In one embodiment, a system is configured to identify, based on predetermined criteria, a first set of users of an online system who belong to a population segment. The system may monitor activities performed by the first set of users on the online system over a predetermined period of time and store the monitored activities as time-series data. A feature set associated with the first set of users may be generated by transforming the time-series data into a frequency domain. The system may train a machine-learning model using the feature set and other feature sets to determine whether activities associated with a given set of users exhibit diurnal behavior pattern. Using the trained machine-learning model, the system may determine whether activities performed by a second set of users on the online system exhibit diurnal behavior pattern.

Abstract: In some aspects, computer-implemented methods of identifying patterns in time-series social-media data. In an embodiment, the method includes applying a deep-learning methodology to the time-series social-media data at a plurality of temporal resolutions to identify patterns that may exist at and across ones of the temporal resolutions. A particular deep-learning methodology that can be used is a recursive convolutional Bayesian model (RCBM) utilizing a special convolutional operator. In some aspects, computer-implemented methods of engineering outcome-dynamics of a dynamic system. In an embodiment, the method includes training a generative model using one or more sets of time-series data and solving an optimization problem composed of a likelihood function of the generative model and a score function reflecting a utility of the dynamic system.